no message

main.py

@@ -20,49 +20,34 @@ nltk.data.path.append(os.getenv('NLTK_DATA'))
 app = FastAPI()
 
 # Initialize the InferenceClient with your model
-client = InferenceClient("mistralai/Mistral-7B-Instruct-v0.2")
-
-
-
-# summarizer = pipeline("summarization", model="sshleifer/distilbart-cnn-12-6")
-
+client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")
 
 class Item(BaseModel):
     prompt: str
     history: list
     system_prompt: str
     temperature: float = 0.8
-    max_new_tokens: int = 12000
+    max_new_tokens: int = 9000
     top_p: float = 0.15
     repetition_penalty: float = 1.0
 
-def format_prompt(current_prompt, history):
-    formatted_history = "<s>"
-    for entry in history:
-        if entry["role"] == "user":
-            formatted_history += f"[USER] {entry['content']} [/USER]"
-        elif entry["role"] == "assistant":
-            formatted_history += f"[ASSISTANT] {entry['content']} [/ASSISTANT]"
-    formatted_history += f"[USER] {current_prompt} [/USER]</s>"
-    return formatted_history
-
+def format_prompt(message, history):
+    prompt = "<s>"
+    for user_prompt, bot_response in history:
+        prompt += f"[INST] {user_prompt} [/INST]"
+        prompt += f" {bot_response}</s> "
+    prompt += f"[INST] {message} [/INST]"
+    return prompt
 
 def generate_stream(item: Item) -> Generator[bytes, None, None]:
     formatted_prompt = format_prompt(f"{item.system_prompt}, {item.prompt}", item.history)
-    # Estimate token count for the formatted_prompt
-    input_token_count = len(nltk.word_tokenize(formatted_prompt))  # NLTK tokenization
-
-    # Ensure total token count doesn't exceed the maximum limit
-    max_tokens_allowed = 32768
-    max_new_tokens_adjusted = max(1, min(item.max_new_tokens, max_tokens_allowed - input_token_count))
-
     generate_kwargs = {
         "temperature": item.temperature,
-        "max_new_tokens": max_new_tokens_adjusted,
+        "max_new_tokens": item.max_new_tokens,
         "top_p": item.top_p,
         "repetition_penalty": item.repetition_penalty,
         "do_sample": True,
-        "seed": 42,
+        "seed": 42,  # Adjust or omit the seed as needed
     }
 
     # Stream the response from the InferenceClient
@@ -74,10 +59,6 @@ def generate_stream(item: Item) -> Generator[bytes, None, None]:
         }
         yield json.dumps(chunk).encode("utf-8") + b"\n"
 
-
-class SummarizeRequest(BaseModel):
-    text: str
-
 @app.post("/generate/")
 async def generate_text(item: Item):
     # Stream response back to the client